Skip to content

Line REPL tool rows: remaining rendering after ADR 0019 #601

Description

@justrach

Problem

ADR 0019 made tool rows one interpreted line instead of live bash / JSON envelopes. That part works:

  ✓ read  src/version.zig  full · 72 lines
  ✓ read  docs/architecture.md  lines · 161 lines

What is still wrong:

  1. mode=lines is tautological. lines · 161 lines prints the CodeDB mode name as if it were a decision. full · 72 lines is only slightly better. The user-relevant fact is architecture.md · 161 lines.
  2. The wall of prose still follows. After those two rows the model dumps a full architecture essay (layout, branch vs main, how the watcher works). The chrome cannot stop model narration; there is no fold/expand on the line REPL, so the transcript still is not something you can stare at all day.
  3. No progressive disclosure. Raw output is /debug or the TUI fold only. Enter/click to expand a one-liner was part of the original sketch and is not built.
  4. JSON interpretation is still mode-leaky. interpretJson surfaces mode verbatim (section, outline, lines, full) instead of a noun the user would miss (161 lines, outline).

Not this issue

Do not bring back live bash dumps, command-as-detail, or the {"ok":true,"file":...} envelope. Those are the regressions ADR 0019 exists to prevent.

Wanted

  ✓ read  version.zig · 72 lines
  ✓ read  architecture.md · 161 lines

and a way to open the raw result on demand. Model essays are a separate prompt-policy question; this issue is the harness chrome.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions